Watkins Alexandra, Ullal Akshith, Sarkar Nilanjan
IEEE Trans Vis Comput Graph. 2024 Apr 17;PP. doi: 10.1109/TVCG.2024.3388376.
User-Avatar interaction within augmented reality applications is rapidly increasing in frequency. Applications routinely place users in rooms with other, remote users embodied by photorealistic avatars, or require users to work with an avatar of a remote user to complete a task. During these types of interactions, it is often required to modify or redirect the posture of an avatar to achieve goals such as contact with or pointing at an object or maintaining eye gaze with the local user. A key limitation of modern redirection techniques is successfully preserving body posture, a critical component of nonverbal communication. This paper presents a new pose-preserving objective function to be used in the multi-objective optimization of an avatar's kinematic configuration. This objective function not only mimics the correct placement of body joints, but also preserves their orientation in space. We have tested this approach against several commonly used and current state-of-the-art redirection techniques and have found that our new approach achieves a significant reduction in targeted redirection error while simultaneously reducing body posture error. Additionally, human subject testing has shown that our new technique provides both a significantly more natural looking redirection and a significantly more realistic and believable overall body posture.
在增强现实应用中，用户与虚拟化身之间的交互频率正在迅速增加。应用程序通常会将用户置于房间中，与由逼真的虚拟化身所代表的其他远程用户在一起，或者要求用户与远程用户的虚拟化身合作完成任务。在这些类型的交互过程中，经常需要修改或重新调整虚拟化身的姿势，以实现诸如与物体接触或指向物体，或与本地用户保持目光接触等目标。现代重定向技术的一个关键限制是能否成功保持身体姿势，而身体姿势是非语言交流的关键组成部分。本文提出了一种新的姿势保持目标函数，用于虚拟化身运动学配置的多目标优化。该目标函数不仅模仿身体关节的正确位置，还能保持它们在空间中的方向。我们已将此方法与几种常用的和当前最先进的重定向技术进行了测试，发现我们的新方法在显著降低目标重定向误差的同时，还能减少身体姿势误差。此外，人体受试者测试表明，我们的新技术既能提供明显更自然的重定向效果，又能呈现出明显更逼真、更可信的整体身体姿势。
